Abstract
To mitigate the annual financial losses caused by SMS phishing (smishing) in South Korea, we propose an explainable smishing detection framework that adapts to a Korean-centric large language model (LLM). Our framework not only classifies smishing attempts but also provides clear explanations, enabling users to identify and understand these threats. This end-to-end solution encompasses data collection, pseudo-label generation, and parameterefficient task adaptation for models with fewer BertScore 0.90 0.68 BLEU 0.70 0.25 0.19 0.12 0.06 0.52 0.35 0.18 Accuracy 1.00 0.75 0.45 0.22 0.50 0.25 0.25 0.50 0.25 ROUGE 1.25 1.25 2.50 3.75 5.00 Logicalness 2.50 1.25 1.25 2.50 2.50 3.75 3.75 F1 (Normal) 1.00 0.75 0.50 3.75 0.75 5.00 1.00 F1 (Smishing) Format Adherence than five billion parameters. Our approach achieves a 15% improvement in accuracy over GPT-4 and generates high-quality explanatory text, as validated by seven automatic metrics and qualitative evaluation, including human assessments.
카카오뱅크 금융기술연구소
Financial Tech Lab